24
Quantitative Marketing and Economics, 1, 179–202, 2003. # 2003 Kluwer Academic Publishers. Manufactured in The Netherlands. The Impact of Frequent Shopper Programs in Grocery Retailing RAJIV LAL* Harvard Business School, Boston, Massachusetts, 02163 E-mail: [email protected] DAVID E. BELL Harvard Business School, Boston, Massachusetts 02163 E-mail: [email protected] Abstract. Frequent shopper programs are becoming ubiquitous in retailing. Retailers seem unsure however about whether these programs are leading to higher loyalty, or to higher profits. In this paper we analyze data from a U.S. supermarket chain that has used a number of frequent shopper rewards to improve sales and profitability. We find that while these programs are profitable, this is only because substantial incremental sales to casual shoppers (cherry pickers) offset subsidies to already loyal customers. In this way our findings are inconsistent with existing theories about how frequent shopper programs are supposed to work. We construct our own Hotelling-like model that explicitly models cherry picking behavior and show that its predictions match the data quite closely. We further test the predictions of our model by characterizing the impact of such programs on trip frequency and basket size. We then use the model to examine more complex scenarios. For example, our analysis suggests that frequent shopper programs may be unprofitable if they eliminate all cherry picking. This may explain why some retailers seem dissatisfied with their programs. We end by proposing a solution that retains the benefits of the frequent shopper programs and yet continues to let supermarkets benefit from price discrimination. Key words. frequent shopper programs, loyalty programs, grocery retailing, game theory JEL Classification: M2, M3, C72 1. Introduction Ever since the acclaimed success of frequent flyer programs in the airline industry, companies in many retail sectors such as hotels, financial services, and grocery, have rushed to introduce ‘‘frequent shopper programs.’’ These programs offer various incentives and rewards to consumers on the basis of cumulative purchases from a given provider, be it a store, a service, or a manufacturer. The simplest frequent shopper program is perhaps a volume discount as in the case of the Discover card where one gets 2% of spending as cash back at the end of the year. More elaborate programs include tiered reward structures backed by a combination of a sort of *Corresponding author.

The Impact of Frequent Shopper Programs in Grocery …people.hbs.edu/dbell/frequent shopper programs.pdf · evenwithendogenousswitchingcostsasinCaminalandMatutes(1990)andKimet al.(2000)

  • Upload
    lamanh

  • View
    217

  • Download
    3

Embed Size (px)

Citation preview

Quantitative Marketing and Economics, 1, 179–202, 2003.

# 2003 Kluwer Academic Publishers. Manufactured in The Netherlands.

The Impact of Frequent Shopper Programs inGrocery Retailing

RAJIV LAL*

Harvard Business School, Boston, Massachusetts, 02163

E-mail: [email protected]

DAVID E. BELL

Harvard Business School, Boston, Massachusetts 02163

E-mail: [email protected]

Abstract. Frequent shopper programs are becoming ubiquitous in retailing. Retailers seem unsure

however about whether these programs are leading to higher loyalty, or to higher profits. In this paper we

analyze data from a U.S. supermarket chain that has used a number of frequent shopper rewards to

improve sales and profitability. We find that while these programs are profitable, this is only because

substantial incremental sales to casual shoppers (cherry pickers) offset subsidies to already loyal

customers. In this way our findings are inconsistent with existing theories about how frequent shopper

programs are supposed to work. We construct our own Hotelling-like model that explicitly models cherry

picking behavior and show that its predictions match the data quite closely. We further test the predictions

of our model by characterizing the impact of such programs on trip frequency and basket size. We then use

the model to examine more complex scenarios. For example, our analysis suggests that frequent shopper

programs may be unprofitable if they eliminate all cherry picking. This may explain why some retailers

seem dissatisfied with their programs. We end by proposing a solution that retains the benefits of the

frequent shopper programs and yet continues to let supermarkets benefit from price discrimination.

Key words. frequent shopper programs, loyalty programs, grocery retailing, game theory

JEL Classification: M2, M3, C72

1. Introduction

Ever since the acclaimed success of frequent flyer programs in the airline industry,companies in many retail sectors such as hotels, financial services, and grocery, haverushed to introduce ‘‘frequent shopper programs.’’ These programs offer variousincentives and rewards to consumers on the basis of cumulative purchases from agiven provider, be it a store, a service, or a manufacturer. The simplest frequentshopper program is perhaps a volume discount as in the case of the Discover cardwhere one gets 2% of spending as cash back at the end of the year. More elaborateprograms include tiered reward structures backed by a combination of a sort of

*Corresponding author.

volume discount (a free ticket for every 25,000 miles), a variety of services and aplethora of direct mail offers with varying degrees of customization.These programs are expensive to launch and maintain. According to Nick

Agarwal, a spokesperson for Asda, U.K., ‘‘it cost us 8 million [pounds] per year justto run the loyalty card trial and would have cost 60 million to roll out nationally(Curtis, October 7 1999).’’ More recently, Starwood Hotels and Resorts WorldwideInc. launched an aggressive frequent-guest program backed by a $50 millionadvertising campaign (The Wall Street Journal, February 2 1999). One of the mostsophisticated programs of this type is being used in the casino industry with Harrah’sinvesting more than $100 million in computers and software to develop andimplement a ‘‘frequent better program.’’ The ‘‘Total Rewards’’ program launched in1997 was recently modified to include gold, platinum and diamond thresholds forfrequent gamblers.While it is difficult to document the economic value of these programs, The Wall

Street Journal (May 5 2000) reports that ‘‘the results are impressive enough thatother casinos are copying some of Harrah’s more discernible methods.’’ In thetheoretical literature, success of these frequent shopper programs is argued to be dueto reduced price competition through the creation of switching costs (Klemperer,1987). The argument is that as consumers continue to dedicate an increasing share oftheir wallet in the category to one brand or to one store, they incur a cost to switch toa different store or brand because of inertia effects. Since in many sectors consumerssubscribe to several frequent shopper programs and regularly buy more than onebrand, one may surmise that these inertia effects may not be significant enough toresult in profitable loyalty programs. Fortunately, these results are shown to persisteven with endogenous switching costs as in Caminal and Matutes (1990) and Kim etal. (2000) where consumers are awarded a cash discount for repeat purchases. In amodel where firms compete over two periods and some consumers (heavy buyers)buy in both periods while others (light buyers) buy in only one period, it is shownthat a cash coupon for purchasing in both periods reduces price competition in thesecond period leading to higher overall profitability. The cash coupon provides anincentive to the heavy buyers to remain loyal in the second period and therefore actsas a switching cost to the consumers. In contrast, the results in Kopalle et al. (1999)suggest that ‘‘reward programs can be viewed as powerful, multi-period pricepromotions.’’Success of loyalty programs can also be argued on the basis of the work by

Heskett et al. (1997) who document the value of focusing attention on the most loyalcustomer. They argue that as customer retention costs are generally lower thancustomer acquisition costs, companies are better off focusing attention on their moreloyal customers especially since the top 20% of customers account for 80% ofrevenues and often more than 100% of profits. Similarly, Dowling and Uncles (1997)conclude that ‘‘ . . . programs must enhance the overall value of the product andservice and motivate loyal buyers to make their next purchase.’’In summary, there are at least two competing reasons for the success of

frequent shopper programs: (a) reduced price competition and therefore higher

180 LAL AND BELL

profits due to switching costs, and (b) reduced marketing expenses by focusingattention on retaining the loyal customers and capturing an increasing share of theirwallet.Few studies shed light on the effectiveness of frequent shopper programs. Dowling

and Uncles (1997) conclude that ‘‘given the popularity of loyalty programs, they aresurprisingly ineffective.’’ Sharp and Sharp (1997) studied the impact of the Fly Buys,Australia’s largest consumer loyalty program covering more than 20% of Australianretail spending and enrolling almost 25% of Australians. Consumers gathered pointsat participating retailers and redeemed them for flights and accommodation. Basedon transaction data from a consumer panel, the authors conclude that only ‘‘two ofthe six loyalty program participants showed substantial excess loyalty deviation,’’but such deviations were also observed for non-members of the loyalty program. Theauthors conclude that they find no evidence to support an increased penetration orpurchase frequency resulting from the incentive effects. A second study investigatedthe profitability of Tesco’s loyalty program using data on market share and share ofcategory requirements during 1994, 1996, and 1997, and concluded that ‘‘on theevidence available, there has been little impact on the share loyalty of individualcustomers so far’’ (East et al., 1998). Hence empirical support for these theoreticalarguments have yet to be provided.The only direct evidence in support of frequent shopper programs is available in

Dreze and Hoch (1998) who experimented with a Baby Bucks program for a periodof six months at all 70 locations of ABCO Markets in Phoenix and Tucson, Arizona.The program, backed by radio and TV commercials, window banners, both in-aisleoverhead banners and shelf talkers, offered consumers Baby Bucks that could beredeemed for a $10 store-wide gift certificate after spending $100 during an earnperiod in the ‘‘baby products’’ category. The program yielded a 25% increase incategory sales, transaction size on baby products also went up by 7.5% and thenumber of customers buying baby products rose by 25% while store traffic increasedby only 5%.In this paper we investigate the effectiveness of frequent shopper programs in

grocery retailing. Grocery retailing is one of the least profitable sectors of theeconomy with net margins of 1–2% and the competition for shoppers is fierce. Thereare not many ways to differentiate grocery stores. Weekly fliers in any givengeographical area, based on the same vendor supported trade promotions, are alsounable to create a point of difference in the eyes of the consumer. Not surprisingly,grocery shoppers have become well known for shopping at more than one storeregularly. Even more importantly, these shoppers are known to split their weeklygrocery baskets across several competing formats. Given the success of frequent flyerprograms in an environment where one airline is perceived to be no different thananother, the grocery industry has moved to adopt the frequent shopper program as apanacea for its misfortunes.Many frequent shopper programs deliver discounts at the check-out and allow

shoppers to earn a rebate for buying a targeted amount of groceries within a welldefined period. As the latest craze in the grocery industry, 61% of retailers had or

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 181

planned to have a frequent shopper program, according to a Food MarketingInstitute survey (The Commercial Appeal, May 17 1998). More recently, RetailAdvertising and Marketing Association International reported that the averagehousehold participating in any frequent shopper program has 3.2 cards (Shepherd-son, February 2000). 66% of U.S. households belong to at least one grocery frequentshopper program and 57% belong to two. However, the success of these loyaltyprograms in the grocery industry is unclear (Kramer, September 15 2000). Kramercontends that even though Grocery Manufacturers of America’s most recent reporton the subject argues that ‘‘leveraging consumer data is critically important to thefuture of the grocery industry, on the basis of Kraft’s and Procter & Gamble’sexperience at Wegmans, Wegmans is the exception rather than the rule. Mostfrequent shopper programs are really frequency programs that use discounts to swayconsumer loyalty. This short term approach eventually becomes just a sophisticatedform of matchable price competition. Most card programs are offered to shoppersindiscriminately and deliver incentives across the board, regardless of shoppervalue.’’ In this paper we provide the first empirical study of the impact of differentimplementations of frequent shopper programs on household shopping behavior ingrocery retailing.The rest of the paper is organized as follows. In the next section we analyze

frequent shopper data that have been made available to us by a grocery chain. Wefind that while the program achieved its objective of increasing spending by its bestcustomers (measured by spending levels), more surprisingly, we find that the increasein spending by customers in lower spending deciles is even greater than that by itsbest customers. In contrast, the redemption of the rewards is the highest among thehigher spending deciles. The chain makes incremental profits from the frequentshopper programs despite the programs’ lack of profitability with the best customers.In Section 3 we construct a model whose predictions match our empirical findings.

Our Hotelling-like one-period model explicitly allows for consumers to cherry-pickbetween competing stores. It predicts that frequent shopper programs improveprofitability when one of the competing stores offers a program. Our model showsthat the increase in profits stems from the reduction in shopping costs motivated bythe rewards of the frequent shopper program. These system wide savings arereflected in higher profits to the stores and better prices to consumers whoincreasingly shop for the basket at a single store. We use data on trip frquency andbasket size to provide empirical support for our theoretical model.In Section 4, we extend our model of frequent shopper programs in two ways.

First, we demonstrate that frequent shopper programs continue to be profitable evenwhen the competing store offers a frequent shopper program. Second, we considersituations where stores sell to consumers with varying shopping costs. We show thatfrequent shopper programs can reduce profitability by denying stores theopportunity to price discriminate between customers. In this more realistic settingwe discover how and why frequent shopper programs can be a money sink forcompeting stores.We conclude with a discussion of our results and directions for future research.

182 LAL AND BELL

2. Frequent shopper program data

We have been provided frequent shopper program data by a supermarket chain inthe mid-west United States that prefers to be anonymous. The chain has a dominantshare of its market. The clientele of this chain tends to be more upscale, older, moreeducated, with higher income and a smaller family. The only major source ofcompetition is from Albertsons which did not have a loyalty card. The chain has hada card program since late 1995 and has used it to provide a variety of benefits to itscard customers in the form of clipless coupons, item discounts, video rentals, rewardsfor loyalty, surprise rewards and automatic contest entries. By 1998, the chain hadabout 200,000 households in the program with about the top 20% of customersaccounting for 80% of sales. Almost 90% of sales and 70% of transactions werereported to be swiped through the card.During 1998 and 1999 the chain implemented a series of reward programs to

increase sales to its best customers. The chain described its best customers interms of spending levels rather than customer profitability. In 1998 the chainintroduced three different kinds of frequent shopper programs. The first, aroundEaster, was a ham promotion where consumers who spent $475 or more in thestore during a six-week earn period received a certificate for a whole ham to beredeemed during a five-week period around Easter. Those spending between$325–$474 received a certificate for a half ham. A second frequent shopperreward program was run during another six-week period during April–May withthe redemption period set for the month of June. In this program, consumers hadto spend $600 or more to get a certificate for a 15% discount coupon for a singlepurchase, a 10% discount coupon was awarded to those spending between $450–$599 and a 5% discount coupon was mailed to those spending between $150–$449. The final frequent shopper program of the year was implemented for aperiod of eight weeks starting soon after Labor Day and ending beforeThanksgiving. Customers spending more than $815 got a certificate for aButterball turkey plus a 15% discount coupon to be redeemed over the followingfour-week period. Customers spending between $625–$815 received a certificatefor a Butterball turkey, and consumers spending between $485–$624 received acertificate for a Little Butterball turkey.In 1999, the supermarket chain did not run a ham promotion but improved the

terms of the 5/10/15% discount to 10/15/20%. This frequent shopper program wasalso implemented during a six-week period during April–May, 1999, with theredemption period set for the month of June. In this program, consumers had tospend $600 or more to get a certificate for a 20% discount coupon for a singlepurchase, a 15% discount coupon was awarded to those spending between $400–$599 and a 10% discount to those spending between $200–$399. The chain also ranan improved turkey promotion in 1999 by decreasing the required spending levels toqualify for a discount coupon. Customers spending more than $750 over a six-weekperiod earned a certificate for a Butterball turkey plus a 15% discount coupon to beredeemed over a three-week period. Customers spending between $500–$749

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 183

received a certificate for a Butterball turkey, and consumers spending between $250–$499 received a certificate for a Little Butterball turkey.

2.1. Analysis of the ham program

As indicated earlier, the objective of the ham program was to increase sales to thebest customers and consequently the profitability of the stores. To analyze theimpact of the ham program and estimate the differential impact on the bestcustomers (high spenders) and the worst customers (low spenders), householdspending data are available to us for the following time periods: a pre-hampromotion period, denoted as P1, lasting from January 1–February 7 1998; the hampromotion period, denoted as P2, during February 8–March 21 1998; and theredemption period, denoted as P3, of March 22–April 18 1998. We also have data forthe corresponding three periods in 1999 when no such ham program was offered.The available data can be summarized as shown in Table 1 with the six differentperiods identified as P1–P6. Moreover, we know that all those who qualified toreceive a reward by spending more than $325 during the earn period were mailed areward to be redeemed during P3. All households who redeemed the award are alsoidentified in our data set.In order to estimate the impact of the ham program on the best and worst

customers, our approach is to look at a pool of best customers (defined later) andestimate the change in spending levels during the ham promotion period ascompared to a control condition. The control condition provides an estimate of thespending level that could be expected of them in the absence of the ham program. Tothe extent that redemption rates and hence involovement in the program may varyacross customer groups, we compare these changes across consumers who redeemedthe reward versus those who did not redeem the reward to identify the impact of theprogram. This analysis is based on the presumption that consumers who are notinterested in the program or do not pay attention to the program, do not changetheir shopping behavior and therefore do not redeem the reward even if they qualify.In contrast, consumers who pay attention to the program are also likely to redeemthe reward. Hence, a comparison of the changes in spending levels during the hampromotion period across these two groups (redeemers and non-redeemers) ofconsumers provides an estimate of the impact of the ham program. Similar estimatesfor the low spenders provide us the variance in the impact of the program across theconsumer population. We recognize that there may be some customers who paid

Table 1. Available data for ham program.

January 1–February 7 February 8–March 21 March 22–April 18

1998 P1-control P2-ham promotion P3-redemption

1999 P4-control P5-control P6-control

184 LAL AND BELL

attention to the frequent shopper program but could not redeem the reward forother extraneous reasons or some who had not heard of the program until theyreceived a coupon in the mail. Such possibilities are likely to result in our estimatesof the impact of the program to be conservative.1

The impact of the frequent shopper program during February 8–March 21 1998can be assessed by comparing the household purchases during the ham promotionperiod with those in a control condition, the pre-ham-promotion period, January 1–February 7 1998.To classify customers as best customers versus worst customers, we use the cut-

offs in the ham promotion program to form natural categories. The ham programrewarded all those spending $475 or more during the promotion period with acoupon for a full ham, those spending between $325 and less than $475 with acoupon for half ham and, those spending less than $325 were not rewarded. Wetherefore categorized all those who were expected to spend $475 or more as the bestcustomers, all those who were expected to spend between $325 and less than $475 asbetter customers and the rest were classified as worst. We used spending levels (S1) inthe pre-ham promotion period in 1998 (P1) to estimate the expected spending levelsduring the ham promotion period and classified consumers as best, better and worst.We can now estimate the impact of the ham promotion program by regressing the

dependent variable, spending level during the ham promotion period (P2) minus thespending level in the pre-ham promotion period (P1) (normalized for the samenumber of days), S2–S1, on the following six independent variables with no constant

1 More precisely, we hypothesize that if a consumer is involved in the program, the impact of the

program as measured by the spending level in the promotion period is

Sjinvolved ¼ Sjcontrol þ effect of the programþ error, and

Skuninvolved ¼ Skcontrol þ error;

where Sj are all the households that are involved in the frequent shopper program and therefore pay

attention and respond to it; Sk are all the households that are not involved in the program, pay no

attention to it and therefore are not affected by it. However, we have no indication of a household’s

level of involvement in our data set. We only observe if the household redeemed the reward that was

mailed to them upon qualifying for one. Therefore,

Slredeemed ¼ Slcontrol þ effect of the program* bþ error, and

Smdid not redeem ¼ Smcontrol þ effect of the program* aþ error;

where b captures the impact of consumers who redeemed the award but did not hear about it and acaptures the fact that some consumers might have been affected by the program but did not redeem the

reward for some extraneous reasons. Thus the difference between those who redeemed the reward and

those who did not redeem the reward provides a conservative estimate of the impact of the frequent

shopper program.

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 185

term in the regression equation. The independent variables in the regression equationare indicator variables for the following categories:

a. Best customers who redeemed the rewardb. Best customers who did not redeem/get the rewardc. Better customers who redeemed the rewardd. Better customers who did not redeem/get the rewarde. Worst customers who redeemed the rewardf. Worst customers who did not redeem/get the reward

The results of this regression analysis are presented in Table 2. These results indicatethat while difference in spending between customers who redeemed the reward andthose who did not redeem the reward among the best customers was $98.02, thecorresponding difference among the better and worst customers was much higher,$140.98 and $150.55, respectively. We therefore reject the null hypothesis andconclude that the frequent shopper program had an impact on shopping behavior.Moreover, we observe that the impact of the program is not the highest among thebest customers. It is the better customers and worst customers who seem to respondmore to the Ham promotion as compared to the best customers. These differencesacross the best, better and worst customers are even more salient when one takes intoaccount the fact that the expected average spending of the best customers was $641while the expected average spending for the better customers was only $387. Withaverage expected spending for all others to be $172, we can conclude that the Hampromotion had the biggest impact in percentage terms on the worst customers andthe least impact on the best customers: contradicting the dictates of Woolf (1996).We also observe that 5941 of the ‘‘best’’ customers redeemed the reward, yielding a

redemption rate of 69.7%. Redemption rates among the ‘‘better’’ and ‘‘worst’’customers were significantly lower at 42.0% and 12.4%, respectively. Clearly, the‘‘worst’’ customer group consists of consumers who were not expected to qualify forthe reward and therefore the low redemption rates are not surprising. However, asalso reported in Table 2, these differences in redemption rates continue to hold evenwhen we reduce the sample to only those consumers who qualified to receive thereward due to their spending in period 2. Thus we notice that while the impact of the

Table 2. The ham program in 1998.

S.E. Variable Coefficient S.E. t-stat R-NR Redemption Redemption (among qualified)

475 R � 35.9 1.78 � 20.2 $98 69.7% 74.2%

475 NR � 133.9 2.7 � 49.7

325 R 75.9 2.77 27.4 $141 42.0% 59.3%

325 NR � 65.1 2.36 � 27.6

Other R 172.6 2.94 58.7 $150.5 12.4% 49.4%

Other NR 22.1 1.11 19.9

Dependent variable, S2–S1. S.E., standard error. t-stat., t-statistic.

186 LAL AND BELL

Ham promotion program is the least among the best customers, the redemption rateis the highest among this group.In order to assess the profitability of this program, transactional level cost data

were needed but were unavailable. However, we are informed by the grocery chainthat the average cost of goods sold is about 75% and that the retail value of full hamand half ham was $25 and $15, respectively. The data show that 5941 consumersincreased their spending by $98.02, 2454 consumers increased their spending by$140.98 and 2170 consumers by $150.55. Moreover, while 4461 consumers redeemedhalf a ham, 6104 consumers redeemed a full ham. Therefore we can calculatethe profitability of the program to be 5941 * $98.02 * 0.25þ 2454 * $140.98 * 0.25 þ2170 * $150.55 * 0.25� 4461 * 0.75 * $15� 6104 * 0.75 * $25 ¼ $149,111.90 from the31,789 consumers, before overhead costs.To summarize, our analysis allows us to conclude that:

i. The ham program had the least impact on the shopping behavior of the bestcustomers.

ii. The percentage of customers redeeming the reward is the highest among the bestcustomers.

iii. The program was profitable not because of the impact on the shopping behaviorof the best customers but instead most of the profitability was due to the impacton the behavior of all but the best customers.

Before accepting these results at their face value we recognize that there may be someconcerns about our estimates. First we need to consider the possibility of astockpiling effect. It is well known that promotions often lead consumers to buymore in the promotion period, only to purchase less in future periods therebycreating the well known post-promotion dip. We need to look at sales in theredemption period and check if they were adversely affected by the purchasingbehavior during the ham promotion period. To investigate this issue, we use(S3� S6), sales in redemption period in 1998 (P3) minus the sales in thecorresponding period in 1999 (P6) (when no ham promotion was offered).(S3� S6) should be negative in the presence of a stockpiling effect. Therefore, thedifference between the estimate of (S3� S6) for those who redeemed the coupon andthose who did not redeem the coupon should also be negative in presence of thestockpiling effect. The results of the regression analysis with (S3� S6) as thedependent variable and the six indicator variables used before as the independentvariables are presented in Table 3. Consumers were classified as best, better andworst on the basis of their spending levels (S1)in the pre-ham-promotion period (P1).The regression results show that the difference between those who redeemed thecoupon and those who did not redeem the coupon continues to remain positive andsignificant. This implies that the ham promotion actually had a carry-over effectrather than suffering from a stockpiling effect. We also ran a similar regression with(S2þ S3)� (S5þ S6) as the dependent variable. (S2þ S3) are the purchases duringand after the ham promotion period in 1998 and (S5þ S6) are purchases in the

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 187

corresponding periods in 1999. If a stockpiling effect was present, we would expectthe difference between those who redeemed the coupon and those who did notredeem the coupon to disappear. However, our results indicate that the difference forthe three groups remain significant and again are the lowest for the best customers asa percent of average expected spending.A second concern relates to possible seasonality in the data, sales closer to Easter

are generally higher than those at the beginning of the year, for example. Wetherefore re-analyze the data controlling for seasonality. We use data in February 8–March 21 1999 as the control condition because the ham program was not offered in1999. While the comparison between purchases during February 8–March 21 1998(ham promotion) and February 8–March 21 1999 (no ham promotion) controls forseasonality, these differences may be affected by year-to-year trends due to inflationor natural changes in family consumption patterns. To control for possible year-to-year trends, we use the difference in sales in Periods 1 (January 1–February 7 1998)and 4 (January 1–February 7 1999) identified in Table 1. Therefore the impact of theham program can be measured as (S2� S5)� (S1� S4). We use spending levels inthe pre-ham promotion period in 1998 (P1) to estimate the expected spending levelsand classify consumers as best, better and worst. Using the same independentvariables as used before, the regression analysis yielded the results presented inTable 4. The results of this regression show that among consumers who wereexpected to spend $475 or more (best customers), the difference in consumerspending between those who redeemed the coupon and those who did not was

Table 3. Checking for stockpiling.

Variable Coefficient S.E. t-stat R-NR

475 R 49.9 1.9 26.3 $57.70

475 NR � 7.8 2.9 � 2.7

325 R 31.6 2.9 10.7 $42

325 NR � 10.4 2.5 � 4.1

Other R 18.4 3.1 5.9 $29.90

Other NR � 11.5 1.2 � 9.7

Dependent variable, S3� S6. S.E., standard error. t-stat., t-statistic.

Table 4. Checking for seasonality and trends.

Variable Coefficient S.E. t-stat R-NR

475 R � 24.76 2.63 � 9.42 $94.46

475 NR � 119.22 3.99 � 29.9

325 R 75.02 4.09 18.33 $138.59

325 NR � 63.57 3.49 � 18.2

Other R 156.64 4.35 36 $144.03

Other NR 12.61 1.64 7.69

Dependent variable: (S2� S5)� (S1�S4), S.E., standard error. t-stat., t-statistic.

188 LAL AND BELL

$94.46, during the ham promotion period. Similarly, among consumers who wereexpected to spend between $325 and less than $475 (better customers), thecorresponding difference was $138.59; and finally for those expected to spend lessthan $325 (worst), the difference was $144.03. Thus the basic findings reportedearlier remain unaffected by potential seasonality and trends in the data.A third concern relates to the possibility of mis-classification of a consumer given

our use of spending levels in the pre-ham promotion period in 1998. In other words,a consumer spending slightly more than $475 is classified as being among the bestcustomers but could actually have been among the better consumers. The spendinglevel used for classification is equal to a true value plus an error term. Someconsumers who are classified as the best consumers may actually be among the betterconsumers, and some consumers who are classified as better consumers may actuallybe among the best consumers. A similar possibility exists among the better and theworst consumers. The question arises if this error in classification creates biases thatdo not allow us to conclude that the impact of the program is the least for the bestcustomers and the most for the worst customers. To address this concern, weconducted a sensitivity analysis on the cut-off values used to classify consumers to beamong the best, better and worst. For example, when we used a cut-off of $500 and$350 in lieu of $475 and $325 to assign consumers into the three groups: best, betterand worst, the difference in the spending levels between consumers who redeemedthe coupon and those who did not, remain unchanged. Table 5a presents the resultsof this sensitivity analysis using different cut-off values with the dependent variableas S2� S1.We also addressed this issue by taking an extreme step of using only those

consumers who actually spent more than $325 during the ham promotion period andtherefore qualified to receive a coupon for half or full ham. With S2� S1 as thedependent variable, we find that among those who were expected to spend more than$475 during the promotion period (best customers), the difference in spendingbetween those who redeemed and did not redeem was $46.64. The correspondingdifference among the better and worst customers was $59.85 and $71.20 respectively.Thus even among those who qualified and received the coupon, the program was theleast effective among the best customers.Second, we ran a regression on change in spending level due to the ham promotion

period, as measured by S2� S5, on spending in P1, whether a household redeemed a

Table 5a. Checking for misclassification. Difference in spending levels of redeemers and non-redeemers.

Cut-off

500 and 350

Cut-off

475 and 325

Cut-off

450 and 300

Cut-off

425 and 275

Actually spent

325 or more

Best $96.34 $98.02 $98.49 $98.35 $46.64

Better $139.85 $140.98 $144.01 $148.27 $59.85

Worst $145.50 $150.55 $155.54 $157.51 $71.20

Dependent variable: S2�S1.

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 189

reward and the interaction between the two aforementioned variables. Theregression results as shown in Table 5b indicate significant effects for each of thethree independent variables and in particular have a negative sign for the interactionvariable; again confirming the finding that the impact of the program is higher atlower levels of spending.In summary, after checking for potential stockpiling effects, seasonality and

trends, and the possibility of mis-classification of customers, we conclude that

i. the best consumers responded the least to the ham promotionii. the best customers are most likely to redeem the rewardiii. overall the ham program was profitable (before overheads).

2.2. Analysis of the discount and turkey program

In analyzing the impact of the 5/10/15% discount program in 1998, and the 10/15/20%

program in 1999, we again used the spending in the discount program period minusthe spending in the previous period, in the same year, as the dependent variable.However, since there are more categories of spending levels than in the ham program,the dependent variable was regressed on eight indicator variables as identified inTable 6a. Table 6a also presents the differences in spending levels of the redeemersand non-redeemers among the different categories for both 1998 and 1999, plus the

Table 5b. Checking for misclassification.

Variable Coefficient S.E. t-stat

Intercept � 37.11 2.07 � 17.9

Redeem indicator 83.03 4.59 18.08

Spend in P1 0.037 0.007 5.25

Interaction � 0.06 0.01 � 5.78

Dependent variable: S2�S5. S.E., standard error. t-stat., t-statistic.

Table 6a. Results from discount programs.

1998 Discounts R-NR* Redemption 1999 Discounts R-NR* Redemption

600 R $57.19 68.66% 600 R $93.11 72.92%

600 NR 600 NR

450 R $80.66 57.47% 400 R $124.67 58.66%

450 NR 400 NR

150 R $72.19 37.42% 200 R $138.68 39.52%

150 NR 200 NR

Other R $101.42 13.69% Other R $182.76 16.96%

Other NR Other NR

*All effects are significant at 1% level.

190 LAL AND BELL

redemption rates, defined by the number of redeemers divided by the number ofhouseholds in the category. We see that the basic pattern in these results is exactly asobserved in the ham program. The best customers are least responsive to the program,as a percentage of their spending levels, but have the highest redemption rates.Finally, using an analysis similar to that reported for the discount programs, theturkey promotion yielded results that are reported in Table 6b; again reconfirming thepattern of effects seen in the other programs.A summary of the effects of these programs is available in Table 7. They show

three systematic effects:

a. These programs have the greatest impact on the behavior of the lower decilecustomers rather than the best customers.

b. The percentage of customers redeeming the reward is highest among thoseconsumers whose behavior is changing the least.

c. The chain loses money on the best customers because a higher fraction redeemthe reward without changing their buying behavior significantly. However, alower redemption rate among the worst consumers along with a much higherincrease in spending leads to higher profitability for the chain. These frequentshopper programs are profitable in the aggregate.

Table 6b. Results from turkey programs.

1999 Turkey R – NR* Redemption

750 R $143.81 82.72%

750 NR

500 R $145.74 61.59%

500 NR

250 R $121.10 34.24%

250 NR

Other R $107.69 13.47%

Other NR

*All effects are significant at 1% level.

Table 7. Summary of our results.

Ham program R-NR Redemption 1998 Discounts R-NR Redemption

475 $98 69.70% 600 $57.19 68.66%

325 $141 42.00% 450 $80.66 57.47%

Other $151 12.40% 150 $72.19 37.43%

Other $101.42 13.69%

1999 Turkey R-NR Redemption 1999 Discounts R-NR Redemption

750 $143.81 82.72% 600 $93.11 72.92%

500 $145.74 61.59% 400 $124.67 58.66%

250 $121.10 34.24% 200 $138.68 39.52%

Other $107.69 13.47% Other $182.76 16.96%

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 191

3. An explanation

Our empirical research suggests that supermarket frequent shopper programs, ascurrently implemented, are an attempt to get customers to spend more at a store inexchange for a discount—be it a ham, turkey or a discount. In this section we seek tooffer an explanation for the impact of such programs; where it is profitable to offer afrequent shopper program if only one of the competing firms offers the program andthat such programs influence the behavior of the worst customers (defined by theirspending level at the store) more than that of the best customers. Our modelingeffort builds on the work of Lal and Matutes (1994) and Lal and Rao (1997) thatmodels competition between two stores where consumers shop for a basket of goods.Our analysis assumes that two supermarkets, A and B, located at the end of a line ofunit length, carry the same assortment of products as reflected in a typical basket ofgoods purchased by shoppers, but perhaps at different prices. For the sake ofexpositional simplicity, the marginal costs of the goods to the stores are assumed tobe zero.Consumers are located uniformly along the line connecting the two stores and

incur a shopping cost c, to and fro, per unit distance. This parameter is ofteninterpreted as the travel cost to a store but more generally as the parameter thatcaptures the degree of differentiation between stores. The latter interpretationinvolves c as the cost associated with the distance between the consumer’s ideal pointand the location of the store in a perceptual space, where the line connecting the twostores is a vector along dimensions differentiating the two stores. Each consumer isassumed to buy the same assortment of goods on a regular basis and the store choicedecision is determined by the prices of the products and the relative convenience ofthe two stores. Consumers have a reservation price v for each good, and choose notto buy if the item is not available at a price below the reservation price. All prices areknown to consumers before making the store choice decision.As mentioned above, stores carry the same assortment of goods but may price

individual items differently. As a practical matter it is difficult to implement a pricingstrategy in which there are no cherry picking opportunities for the dedicatedshopper. With thousands of items in the grocery store there will always be someitems that are attractive to cherry pickers. The set of all items in each store can bedivided into two baskets: one that consists of items that are relatively cheaper in thestore and the other of those items that are relatively cheaper in the other store. Theprice of the basket that is cheaper in store A is assumed to be Pa � d and the price ofthe basket of goods that is more expensive in store A is assumed to be Pa; similarly,the prices at store B are Pb � d and Pb for these two baskets respectively. We furtherassume that the discount d on the cheaper items in the store is exogenouslydetermined by factors beyond the control of the store and therefore, stores set pricesof the items in the grocery basket so as to communicate a price image captured bythe decision variables Pa and Pb respectively. Since, all prices are known to theconsumers, the cherry picker therefore can pay 2Pa � d by shopping for all items atstore A or pay Pa þ Pb � 2d by cherry picking. Our model therefore has two key

192 LAL AND BELL

parameters that capture the trade off faced by the cherry picker: c, the opportunitycost of shopping around, and d, the savings potential.

3.1. Analysis of the pure price game

For benchmark purposes, we begin by ignoring discounts and setting d ¼ 0. In thiscase there is no incentive for any customer to cherry pick and every consumer shopseither at store A, or store B, but not both. The marginal customer is located at adistance x from store A where

2Pa þ 2cx ¼ 2Pb þ 2cð1� xÞ

i:e: x ¼ cþ Pb � Pa

2c:

Assuming that the reservation price is sufficiently high to cover the price of the goodsand the shopping costs, store A’s profits are

2Pax ¼ Pacþ Pb � Pa

2c

� �;

and are maximized when Pa ¼ 0:5ðcþ PbÞ. By optimizing 2Pbð1� xÞ or simply byrecognizing the symmetry between A and B we deduce that P�

a ¼ P�b ¼ c. Total

profits to store A are 2xPa ¼ c. Note that the profits to the stores are related to travelcosts rather than the value added of the items.These results are well established but serve as a benchmark for our later analyses.

3.2. The case of promotional pricing

If d > 0, consumers that are located closer to store A are likely to purchase thebasket at store A and this fraction is

xa ¼2cþ Pb � d � Pa

2c

� �:

Similarly all consumers located at a distance greater than xb from store A, wherexb ¼ Pb � Pa þ d=2cf g buy the complete assortment from store B. All consumerslocated between xa and xb cherry pick and buy the lower priced good at each store.Such cherry pickers exist only if xb > xa, i.e., d > c, that is the discounts have to belarge enough (compared to the shopping costs) to make cherry picking worthwhile.If the discounts are too large, everyone cherry picks; i.e., xa < 0 and xb > 1. In

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 193

equilibrium this happens if d ¼ 2c, and discounts are large enough to cover the totalcosts of shopping around for all consumers.We conclude that promotions d induce

i. no cherry picking if d c,ii. some cherry picking if 2c d c.

We know that store profitability is equal to c in case (i). What about the other case ?In case (ii), store A will wish to maximize

ð2Pa � dÞxa þ ðPa � dÞðxb � xaÞ ¼ ð2Pa � dÞ 2cþ Pb � d � Pa

2c

þ ðPa � dÞfd=c� 1g:

Differentiating with respect to Pa yields

2þ 1

cfPb � Pa � d � Pa þ 0:5d þ dg � 1 ¼ 1þ Pb

c� 2Pa

cþ d

2c:

Knowing that a symmetric equilibrium dictates P�a ¼ P�

b we see that the optimalsolution is given by

1þ P�a

c� 2P�

a

cþ d

2c¼ 0; i.e. P�

a ¼ P�b ¼ cþ 0:5d:

The equilibrium profits therefore are c� 0:5dðd=c� 1Þ which is less than c for alld > c, and are zero when d ¼ 2c.

3.3. Introducing loyalty programs

As the results in Section 2 are for a mid-west supermarket chain that is the onlycompetitor in its trading area to offer a frequent shopper program, we begin byassuming that only one store, store A, offers a loyalty program. This store offers adiscount of L to all consumers who buy both goods at store A, otherwise, the regularprice at store A is Pa and the discounted price is Pa � d. The profits to store A, ifsome consumers continue to cherry pick, which requires cþ 0:5L < d < 2c, are:

Pa ¼ ð2Pa � d � LÞ Pb � Pa � d þ Lþ 2c

2c

� �þ ðPa � dÞ 2d � 2c� L

2c

� �:

194 LAL AND BELL

Differentiating with respect to L and Pa respectively, we get

qPa

qL¼ 2Pa � 2L� 2c� Pb þ d

2c;

qPa

qPa¼ � 4Pa þ 2Lþ 2cþ 2Pb þ d

2c:

and

Setting the first order conditions to zero and solving, we get

Pa ¼ d þ 0:5Pb and L ¼ 1:5d � c:

At the optimal value of L there is no cherry picking because 2d � 2c� L < 0 whenL ¼ 1:5d � c, and d < 2c. The profits to the stores are c, and are higher than in thecase where no store offers a frequent shopper program.2 These results are consistentwith the empirical results presented in Section 2 where frequent shopper programsare observed to have increased profitability. Our model suggests that the increase inprofitability results from the reduction in the welfare loss incurred due to cherrypicking. This welfare loss is recovered through the frequent shopper program andpart of it goes to the store in the form of increased profitability. Moreover, since thefraction of consumers shopping for a single item is reduced from ð2d � 2cÞ=2c toð2d � 2c� LÞ=2c, it is easily seen that cherry pickers are spending more at store A bybuying the complete basket at the store. This contrasts with the behavior of loyalcustomers which remains unaffected.To summarize, we conclude from our model that:

a. the frequent shopper programs are profitable even when only one retailer offerssuch a program;

b. they are profitable because they help reduce cherry picking and the reduction insystemwide shopping costs show up in the form of higher profits, and

c. the frequent shopper programs impact the behavior of the cherry pickers morethan that of the loyal customers.

2 In case of no cherry picking, the market share of store A is x where x is such that

2Pa � d � Lþ 2cx ¼ 2Pb � d þ 2cð1� xÞ. The profits to store A therefore are ð2Pa � d � LÞ * x andthose to store B are ð2Pb � dÞð1� xÞ. Let 2Pa � d � L ¼ Ya and 2Pb � d ¼ Yb. The profits to store A

can rewritten as Ya *x and those to store B as Yb * ð1� xÞ and x ¼ 2c� Ya þ Yb=4c. Now it easily

seen that in equilibrium Ya ¼ Yb and the profits to the two stores are c.

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 195

3.4. Empirical validity of our model

Assuming that frequent shopper programs do not affect the total number of tripsmade by a households to grocery stores, we are able to empirically test theconclusions of our model by estimating the difference in the number of trips for theredeemers and non-redeemers across the three customer groups. Given ourcontention that frequent shopper programs impact the behavior of cherry pickersmore than that of the loyal customers by inducing them to shop more at the store, weexpect to observe the impact of the program to be the highest for the worstcustomers. Table 8a presents the results of a regression analysis where the dependentvariable, T2�T1, is the difference in the number of trips during the ham promotionperiod and the pre-ham promotion period. As expected, the difference between theredeemers and non-redeemers is the highest among the worst customers; during theham promotion period, among the worst customers, the redeemers made 1.7 tripsmore than the non-redeemers (compared to an average of 5.8 trips for the worstcustomers during the ham promotion period). Similarly, among the best customers,the redeemers made 1.2 trips more than the non-redeemers but the average numberof trips for the best customers was approximately 10 trips during this period. Theaverage number of trips for the better customers was 7.8.Another test of our model could be with respect to the basket size measured as the

spending per trip. Our model predicts that the introduction of the frequent shopperprogram would lead to more consumers buying the basket rather than the singleitem. We should therefore expect the difference in basket size for the redeemers andnon-redeemers to be higher for the worst customers. Table 8b presents the results ofsuch an analysis and shows that while the difference between the redeemers and non-redeemers in the average basket size for the best customers was $5.67, thecorresponding difference for the better and worst customers was $7.56 and $9.33,respectively. Hence the model predictions are again supported by our data. Theseresults find further support in the analysis of the discount program, see Table 8c,where the impact of the program on trip frequency and basket size for the worstcustomers is again more than it is for the best customers.While our theoretical model therefore provides an explanation for the empirical

results presented in Section 2, our experience from talking to retailers that have

Table 8a. Results from ham programs. Differences in trips between redeemers and

non-redeemers.

Variable Coefficient t-statistic R-NR

475 R � 0.1 � 3.7 1.2

475 NR � 1.3 � 21.3

325 R 0.9 14.3 1.6

325 NR � 0.7 � 14.5

Other R 2.2 34.8 1.7

Other NR 0.5 21.7

196 LAL AND BELL

implemented frequent shopper programs suggests that few think they have enhancedprofitability: some customers may indeed be more loyal in the sense of buying agreater percentage of their food items at a single store; but the fruits of this loyaltyhave yet to find their way to the bottom line. Retailers we have talked to that have sofar resisted the introduction of loyalty programs, express satisfaction at this mixedbag performance, since it relieves them of the burden—and potential distraction—ofintroducing their own loyalty/frequent shopper program.As we contrast our empirical findings that paint a positive picture on the success

of frequent shopper programs with the anecdotal evidence suggesting the lack ofenthusiasm for these programs among leading retailers, we note two differencesbetween the circumstances for our mid-western chain and these leading retailers.First, as mentioned earlier, the mid-western chain, unlike many supermarketchains, does not face competing loyalty programs. Second, while our modelassumes homogeneity (every customer has the same travel costs), supermarketsattract consumers with different shopping costs. In the next section we thereforeextend our analysis to investigate the impact of these factors on our theoreticalresults. These extensions will allow us to formulate hypotheses that could be testedin future.

Table 8b. Results from ham programs. Differences in basket size between redeemers and non-redeemers.

Variable Coefficient t-statistic R-NR

475 R � 3.21 � 11.4 $5.67

475 NR � 8.88 � 20.7

325 R 1.67 3.8 $7.56

325 NR � 5.89 � 15.7

Other R 9.66 20.7 $9.33

Other NR 0.33 1.87

Table 8c. Results from discount programs. Differences in trips and basket size between redeemers and

non-redeemers.

R-NR* R-NR*

1998 discounts Trips Basket size 1999 discounts Trips Basket size

600 0.88 $0.66 600 1.08 5.14

450 0.95 $3.15 400 1.51 8.42

150 1.22 $3.71 200 1.84 7.45

Other 1.4 $9.21 Other 2.18 15.56

* All effects are significant at 1% level.

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 197

4. Extending the model

4.1. Competing loyalty programs

We now extend our analysis to the case where both stores offer a frequent shopperprogram. Suppose stores A and B offer loyalty payments of La and Lb to customerswho buy both baskets at their store. Cherry pickers do not receive this payment ofcourse. All consumers located to the left of ya buy the complete assortment fromstore A where ya is such that

2Pa � d � La þ 2cya ¼ Pa � d þ Pb � d þ 2c;

or

ya ¼2cþ Pb � d � Pa þ La

2c

� �:

Similarly, all consumers located to the right of yb buy the complete assortment atstore B where yb is such that

2Pb � d � Lb þ 2cð1� ybÞ ¼ Pa � d þ Pb � d þ 2c;

or

yb ¼Pb þ d � Pa � Lb

2c

� �:

Assuming that some consumers continue to cherry pick, store A would maximizeð2Pa � d � LaÞya þ ðPa � dÞðyb � yaÞ. Optimizing with respect to Pa and La yieldsP�a ¼ 2c=3þ d and L�

a ¼ d � 2c=3 and similarly for P�b and L

�b. These values lead to

y�a ¼ 2=3 and y�b ¼ 1=3. Since y�a > y�b, we conclude that offering these loyaltyprograms eliminates all cherry picking. Any solution with the property P�

a � L�a ¼

cþ d=2 and L�a d is therefore optimal. With the elimination of cherry picking,

store A’s profits are as before. The solution P�a ¼ cþ d;L�

a ¼ d means that the twobaskets are priced at cþ d and c respectively with a loyalty payment of d, for a totalpayment of 2c and profits of c. Thus we find that if loyalty programs are offered byboth stores they do not enhance the profitability of the stores compared to the casewhen only one store offers a loyalty program.

4.2. The case of two customer segments

We will now repeat the analysis of Section 4 assuming that there are two kinds ofcustomers, with shopping costs c1 and c2 respectively where c2 < c1. For the sake ofexpositional simplicity, we assume that there are equal numbers of each type, each

198 LAL AND BELL

distributed uniformly between the stores. To keep the profitability figurescomparable with our earlier results, we assume that the total number of customersis unchanged. As in the previous section, we have a number of cases to consider todetermine the equilibrium prices and profitability to the stores.First consider the case d < c2: In this case no cherry picking takes place and there

are two breakeven distances to calculate, x1 and x2 where, for example,2Pa þ 2c1x1 ¼ 2Pb þ 2c1ð1� x1Þ, and store A will pick Pa to maximize2Paðx1 þ x2Þ=2, the divisor reflecting the density of customers. The optimal solution is

P�a ¼

2c1c2

ðc1 þ c2Þ;

with a profit of 2c1c2=ðc1 þ c2Þ:Next consider the case c2 < d < 2c2 < c1: If d is in this range, some of the c2

customers cherry pick but none of the c1 customers do. We find that the equilibriumprices are

P�a ¼ P�

b ¼2c1c2

ðc1 þ c2Þþ d

2

with store profits of 2c1c2=ðc1 þ c2Þ � dðd=c2 � 1Þ=4:If 2c2 < d < c1, all of the c2 customers cherry pick, but none of the c1 customers

do so. The profits to store A are ð2Pa � dÞðc1 þ Pb � PaÞ=2c1 þ ðPa � dÞ and theequilibrium prices are P�

a ¼ P�b ¼ 2c1 þ d=2 with store profits of 2c1 � d=4:

Finally, when c1 < d < 2c1, even some of the c1 customers cherry pick. Theoptimal price remains P�

a ¼ P�b ¼ 2c1 þ d=2 but store profits are 2c1 � d2=4c1: Figure

1 shows a graph of store profits as a function of d. The highest store profits occurwhen d ¼ 2c2: In other words, store profits are maximized when one segment ofconsumers cherry picks between the two stores.

Figure 1. Store profits as d varies.

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 199

The intuition for this graph is as follows. So long as d < c2, no customer cherrypicks therefore profits are constant for all d c2: When d exceeds c2, somecustomers with lower shopping cost begin to cherry pick resulting in lower profits forthe stores. This continues until d ¼ 2c2 at which point all customers with lowershopping costs are cherry picking. When d > 2c2, profits increase substantially butthen decline for the same reason as before (losses from cherry picking). The keyquestion that remains to be answered is why profits increase so substantially whend > 2c2: The answer is that with d 2c2, some cherry pickers remain in play aspotential full basket shoppers and hence influence full basket prices. However, whend > 2c2, all consumers with shopping costs equal to c2 will never buy both baskets ateither store; as a result the full basket prices are influenced only by the pricesensitivity of the remaining high shopping cost customers.We now consider the impact of competing loyalty programs on the shopping

behavior of the two segments of customers in our model. As before, we can show thatoptimal loyalty programs eliminate cherry picking by setting the reward for buyingboth baskets at the same store to be at or above d.When 2c2 > d > c2, these optimalloyalty programs increase profits to the stores because eliminating cherry pickingrestores profits to the same level as when d ¼ 0 (see Figure 1). However, when d > 2c2,these loyalty programs decrease profits because profits when d ¼ 0 are lower. Thus wesee that introducing competing loyalty programs do not always result in higher profits.The latest evidence to support the lack of performance comes from the UK with

the scrapping of the pilot program by Asda in August 1999 and that by Safeway inMay 2000. The experience at Safeway is summarized by its communications director,‘‘Loyalty is not synonymous with having a loyalty card. Since we scrapped our ABCcard we have gone on gaining customers—one million more have come to us the pastyear. You can give them a point for every penny spent, but it doesn’t buy youloyalty.’’ Similarly, the Asda spokesman said that given the high costs ofimplementing these programs, ‘‘we decided we didn’t have to invest in points andplastic to make our customers loyal.’’ Our results are also consistent with thesentiments expressed by Kramer (2000), where he concludes that ‘‘ . . . most frequentshopper programs are really frequency programs that use discounts to swayconsumer loyalty. This short term approach eventually becomes just a sophisticatedform of matchable price competition. . . . ’’

5. Summary and discussion

The current thinking on frequent shopper programs is that they should be designedto dig deeper into the pockets of the best customers. Our data from a supermarketchain tells a different story. The frequent shopper programs implemented by thesupermarket chain we studied affected the behavior of lower spending customersmore than that of its best customers. Moreover, these programs were profitabledespite the fact that the best customers were more likely to redeem the rewardwithout significant change in their behavior.

200 LAL AND BELL

These empirical results are not surprising considering that cherry picking is, on theface of it, a bad thing. Customers travel more than is strictly necessary; stores sellonly low margin items to these customers. It would seem that a loyalty program thatencourages shoppers to buy all of their goods at a single store can be beneficial to all.We construct a model of competition among stores that highlights the role of the

cherry picker. We capture the essence of their behavior through two parameters:shopping costs ðcÞ and the potential savings from cherry picking ðd Þ. The analysis ofour model makes predictions that are consistent with our empirical findings.Furthermore, we provide empirical support for our model by developing and testingtwo new hypotheses with respect to trip frequency and basket size. Extending themodel to allow for the existence of customers with varying shopping costs shows thatfrequent shopper programs that reduce cherry picking may actually reduceprofitability. This is because price promotions add value by providing a way forsupermarkets to price segment the market—low prices to the price sensitive, high(not so low) prices to the price insensitive. The problem with discount-based frequentshopper programs is that they are fairly costly and offer the same deals to all, thusremoving the hidden benefit of price segmentation. This may explain why groceryretailers seem largely dissatisfied with frequent shopper programs.How can frequent shopper programs be successful in the grocery industry? First of

all a frequent shopper program cannot be offered indiscriminately. It makes littlesense to provide the person who lives next to the supermarket with a discount forloyalty. Yet it would be a pity to lose any of one’s 100 best customers for the sake ofa few dollars—companies often give discounts to their best customers. Our analysis,throughout, assumed undifferentiated stores. The secret seems to be to continue tooffer promotions for the price sensitive cherry picker and value added services for thefull basket, less price sensitive shopper. Retailers can also draw a lesson from thesuccess of loyalty programs in the airline industry. Airlines reward their bestcustomers through improved perks and services rather than better prices anddiscounts even when the cost of offering a free trip is very low. Supermarkets alsoneed to find ways to reward desirable customers with non-price benefits.A final word needs to be added about the ability of such programs to modify

consumer behavior. These awards should be targeted towards consumers who havethe potential to modify their behavior in a way that is desirable to the franchise. Indifferent industries, different type of customers may have the flexibility to changetheir behavior in response to frequent shopper programs. For example, in the airlineindustry, it is the frequent traveller, business customer, who might have thepossibility to increase its patronage of one airline over the other. The infrequenttraveller who often takes only two family vacations a year may not have muchflexibility in choice of airlines and may take a long time to accumulate the miles forfree trips. Hence, it is easier to induce behavior modification among the morefrequent traveller than the infrequent traveller. Frequent reward programs shouldtherefore be directed towards the business customers and rewards have to bedesigned so as to encourage such customers to modify their behavior such that theyare beneficial to the airline. In contrast, in the grocery industry, the best customer

THE IMPACT OF FREQUENT SHOPPER PROGRAMS IN GROCERY RETAILING 201

may already be patronizing the store to a degree that it is difficult to increase thestore’s share of wallet. It may therefore be difficult to significantly change his/herbehavior through a frequent shopper program. Hence rewards to frequent shoppershave to be either lucrative enough to change the behavior of such customers or needto be targeted at a different customer segment.

Acknowledgments

David E. Bell is the George M. Moffett Professor of Agriculture and Business, andRajiv Lal is the Stanley Roth Sr. Professor of Retailing, both at the HarvardBusiness School, Boston, MA, 02163. The authors wish to thank Professors AnandBodapati, Bharat Anand, Alvin Silk and participants of the marketing seminars atHarvard Business School, Simon School of Business, University of Rochester, SternSchool of Management, NYU, University of Virginia and Wharton School ofManagement, for their many helpful comments and suggestions.

References

Caminal, Ramon and Carmen Matutes. (1990). ‘‘Endogenous Switching Costs in a Duopoly Model’’,

International Journal of Industrial Organization 8, 353–373.

Curtis, James. (1999). ‘‘Cards versus Cuts in the Loyalty Wars’’, Marketing, London.

Dowling, Grahame R. and Mark Uncles. (1997). ‘‘Do Customer Loyalty Programs Really Work?’’, Sloan

Management Review Summer, 71–82.

Dreze, Xavier and Stephen Hoch. (1998). ‘‘Exploiting the Installed Base Using Cross-merchandising and

Category Destination Programs’’, International Journal of Research in Marketing 15, 459–471.

East, Robert, Annik Hogg, and Wendy Lomax. (1998). ‘‘The Future of Loyalty Schemes’’, Journal of

Targeting, Measurement and Analysis for Marketing 7(1), 11–21.

Heskett, James L., Earl Sasser, W. Jr., and Leonard A. Schlesinger. (1997). The Service Profit Chain. New

York: Simon and Schuster.

Kim, Byung-Do, Mengze Shi, and Kannan Srinivasan. (2001). ‘‘Reward Programs and Tacit Collusion’’,

Marketing Science 20(2), 99–120.

Kim, Byung-Do, Mengze Shi, and Kannan Srinivasan. (2001). ‘‘Managing Capacity through Rewards

Programs’’. Working paper, Carnegie Mellon University.

Klemperer, Paul. (1987). ‘‘The Competitveness of Markets with Switching Costs’’, Rand Journal of

Economics 18(1) (Spring), 138–150.

Kopalle, Praveen, et al. (1999). ‘‘The Economic Viability of Frequency Reward Programs in a Strategic

Competitive Environment’’. Working paper, Amos Tuck School of Business Administration,

Dartmouth College, Hanover, NH.

Kramer, Jon. (2000). ‘‘Fool’s Gold’’, Supermarket Business. New York.

Sharp, Byron and Anne Sharp. (1997). ‘‘Loyalty Programs and their Impact on Repeat-purchase Loyalty

Patterns’’, International Journal of Research in Marketing 14, 473–486.

Shepherdson, Nancy. (2000). ‘‘Holding all the Cards’’, American Demographics 22(2), 35–37.

Woolf, Brian P. (1996). Customer Specific Marketing. Teal Books, Greenville, SC, USA.

202 LAL AND BELL